DeepSeek Releases V4 Pro and V4 Flash Models with Agentic Improvements and Huawei Chip Support

DeepSeek, the Chinese AI startup that disrupted global markets in 2025, has launched preview versions of its V4 Pro and V4 Flash models, a major update to its open-source AI lineup. The V4 Pro Max achieves 'superior performance' on standard reasoning benchmarks compared to OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro, falling just short of GPT-5.4 and Gemini 3.1-Pro. On agentic capabilities—autonomous task execution—the pro version outperforms Claude's Sonnet 4.5 and approaches Opus 4.5, based on DeepSeek's own evaluations. The flash version matches the pro on simple agent tasks and closely trails in reasoning. Both models feature a 1 million token context window, up from 128K in V3, and support Huawei's Ascend chips, reducing dependence on Nvidia.

The release comes amid escalating US-China AI rivalry, with OpenAI dropping GPT-5.5 just hours earlier. DeepSeek's open-source approach, offering free web and mobile chatbots, contrasts with proprietary US models. While analysts like Omdia's Lian Jye Su deem V4 'very competitive,' Morningstar's Ivan Su cautions it's not as groundbreaking as the R1 reasoning model from January 2025. Huawei confirmed Ascend chip compatibility, marking a step toward technical decoupling from Nvidia's ecosystem. DeepSeek faces accusations from Anthropic and OpenAI of building on their technology, but the V4 rollout underscores China's push for AI self-reliance amid intensifying global competition.